| 1 | /* |
| 2 | * linux/fs/super.c |
| 3 | * |
| 4 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 5 | * |
| 6 | * super.c contains code to handle: - mount structures |
| 7 | * - super-block tables |
| 8 | * - filesystem drivers list |
| 9 | * - mount system call |
| 10 | * - umount system call |
| 11 | * - ustat system call |
| 12 | * |
| 13 | * GK 2/5/95 - Changed to support mounting the root fs via NFS |
| 14 | * |
| 15 | * Added kerneld support: Jacques Gelinas and Bjorn Ekwall |
| 16 | * Added change_root: Werner Almesberger & Hans Lermen, Feb '96 |
| 17 | * Added options to /proc/mounts: |
| 18 | * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996. |
| 19 | * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998 |
| 20 | * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000 |
| 21 | */ |
| 22 | |
| 23 | #include <linux/module.h> |
| 24 | #include <linux/slab.h> |
| 25 | #include <linux/init.h> |
| 26 | #include <linux/smp_lock.h> |
| 27 | #include <linux/acct.h> |
| 28 | #include <linux/blkdev.h> |
| 29 | #include <linux/quotaops.h> |
| 30 | #include <linux/namei.h> |
| 31 | #include <linux/mount.h> |
| 32 | #include <linux/security.h> |
| 33 | #include <linux/syscalls.h> |
| 34 | #include <linux/vfs.h> |
| 35 | #include <linux/writeback.h> /* for the emergency remount stuff */ |
| 36 | #include <linux/idr.h> |
| 37 | #include <linux/kobject.h> |
| 38 | #include <linux/mutex.h> |
| 39 | #include <linux/file.h> |
| 40 | #include <linux/backing-dev.h> |
| 41 | #include <asm/uaccess.h> |
| 42 | #include "internal.h" |
| 43 | |
| 44 | |
| 45 | LIST_HEAD(super_blocks); |
| 46 | DEFINE_SPINLOCK(sb_lock); |
| 47 | |
| 48 | /** |
| 49 | * alloc_super - create new superblock |
| 50 | * @type: filesystem type superblock should belong to |
| 51 | * |
| 52 | * Allocates and initializes a new &struct super_block. alloc_super() |
| 53 | * returns a pointer new superblock or %NULL if allocation had failed. |
| 54 | */ |
| 55 | static struct super_block *alloc_super(struct file_system_type *type) |
| 56 | { |
| 57 | struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER); |
| 58 | static const struct super_operations default_op; |
| 59 | |
| 60 | if (s) { |
| 61 | if (security_sb_alloc(s)) { |
| 62 | kfree(s); |
| 63 | s = NULL; |
| 64 | goto out; |
| 65 | } |
| 66 | INIT_LIST_HEAD(&s->s_files); |
| 67 | INIT_LIST_HEAD(&s->s_instances); |
| 68 | INIT_HLIST_HEAD(&s->s_anon); |
| 69 | INIT_LIST_HEAD(&s->s_inodes); |
| 70 | INIT_LIST_HEAD(&s->s_dentry_lru); |
| 71 | init_rwsem(&s->s_umount); |
| 72 | mutex_init(&s->s_lock); |
| 73 | lockdep_set_class(&s->s_umount, &type->s_umount_key); |
| 74 | /* |
| 75 | * The locking rules for s_lock are up to the |
| 76 | * filesystem. For example ext3fs has different |
| 77 | * lock ordering than usbfs: |
| 78 | */ |
| 79 | lockdep_set_class(&s->s_lock, &type->s_lock_key); |
| 80 | /* |
| 81 | * sget() can have s_umount recursion. |
| 82 | * |
| 83 | * When it cannot find a suitable sb, it allocates a new |
| 84 | * one (this one), and tries again to find a suitable old |
| 85 | * one. |
| 86 | * |
| 87 | * In case that succeeds, it will acquire the s_umount |
| 88 | * lock of the old one. Since these are clearly distrinct |
| 89 | * locks, and this object isn't exposed yet, there's no |
| 90 | * risk of deadlocks. |
| 91 | * |
| 92 | * Annotate this by putting this lock in a different |
| 93 | * subclass. |
| 94 | */ |
| 95 | down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING); |
| 96 | s->s_count = 1; |
| 97 | atomic_set(&s->s_active, 1); |
| 98 | mutex_init(&s->s_vfs_rename_mutex); |
| 99 | mutex_init(&s->s_dquot.dqio_mutex); |
| 100 | mutex_init(&s->s_dquot.dqonoff_mutex); |
| 101 | init_rwsem(&s->s_dquot.dqptr_sem); |
| 102 | init_waitqueue_head(&s->s_wait_unfrozen); |
| 103 | s->s_maxbytes = MAX_NON_LFS; |
| 104 | s->dq_op = sb_dquot_ops; |
| 105 | s->s_qcop = sb_quotactl_ops; |
| 106 | s->s_op = &default_op; |
| 107 | s->s_time_gran = 1000000000; |
| 108 | } |
| 109 | out: |
| 110 | return s; |
| 111 | } |
| 112 | |
| 113 | /** |
| 114 | * destroy_super - frees a superblock |
| 115 | * @s: superblock to free |
| 116 | * |
| 117 | * Frees a superblock. |
| 118 | */ |
| 119 | static inline void destroy_super(struct super_block *s) |
| 120 | { |
| 121 | security_sb_free(s); |
| 122 | kfree(s->s_subtype); |
| 123 | kfree(s->s_options); |
| 124 | kfree(s); |
| 125 | } |
| 126 | |
| 127 | /* Superblock refcounting */ |
| 128 | |
| 129 | /* |
| 130 | * Drop a superblock's refcount. Returns non-zero if the superblock was |
| 131 | * destroyed. The caller must hold sb_lock. |
| 132 | */ |
| 133 | static int __put_super(struct super_block *sb) |
| 134 | { |
| 135 | int ret = 0; |
| 136 | |
| 137 | if (!--sb->s_count) { |
| 138 | destroy_super(sb); |
| 139 | ret = 1; |
| 140 | } |
| 141 | return ret; |
| 142 | } |
| 143 | |
| 144 | /* |
| 145 | * Drop a superblock's refcount. |
| 146 | * Returns non-zero if the superblock is about to be destroyed and |
| 147 | * at least is already removed from super_blocks list, so if we are |
| 148 | * making a loop through super blocks then we need to restart. |
| 149 | * The caller must hold sb_lock. |
| 150 | */ |
| 151 | int __put_super_and_need_restart(struct super_block *sb) |
| 152 | { |
| 153 | /* check for race with generic_shutdown_super() */ |
| 154 | if (list_empty(&sb->s_list)) { |
| 155 | /* super block is removed, need to restart... */ |
| 156 | __put_super(sb); |
| 157 | return 1; |
| 158 | } |
| 159 | /* can't be the last, since s_list is still in use */ |
| 160 | sb->s_count--; |
| 161 | BUG_ON(sb->s_count == 0); |
| 162 | return 0; |
| 163 | } |
| 164 | |
| 165 | /** |
| 166 | * put_super - drop a temporary reference to superblock |
| 167 | * @sb: superblock in question |
| 168 | * |
| 169 | * Drops a temporary reference, frees superblock if there's no |
| 170 | * references left. |
| 171 | */ |
| 172 | void put_super(struct super_block *sb) |
| 173 | { |
| 174 | spin_lock(&sb_lock); |
| 175 | __put_super(sb); |
| 176 | spin_unlock(&sb_lock); |
| 177 | } |
| 178 | |
| 179 | |
| 180 | /** |
| 181 | * deactivate_locked_super - drop an active reference to superblock |
| 182 | * @s: superblock to deactivate |
| 183 | * |
| 184 | * Drops an active reference to superblock, converting it into a temprory |
| 185 | * one if there is no other active references left. In that case we |
| 186 | * tell fs driver to shut it down and drop the temporary reference we |
| 187 | * had just acquired. |
| 188 | * |
| 189 | * Caller holds exclusive lock on superblock; that lock is released. |
| 190 | */ |
| 191 | void deactivate_locked_super(struct super_block *s) |
| 192 | { |
| 193 | struct file_system_type *fs = s->s_type; |
| 194 | if (atomic_dec_and_test(&s->s_active)) { |
| 195 | vfs_dq_off(s, 0); |
| 196 | fs->kill_sb(s); |
| 197 | put_filesystem(fs); |
| 198 | put_super(s); |
| 199 | } else { |
| 200 | up_write(&s->s_umount); |
| 201 | } |
| 202 | } |
| 203 | |
| 204 | EXPORT_SYMBOL(deactivate_locked_super); |
| 205 | |
| 206 | /** |
| 207 | * deactivate_super - drop an active reference to superblock |
| 208 | * @s: superblock to deactivate |
| 209 | * |
| 210 | * Variant of deactivate_locked_super(), except that superblock is *not* |
| 211 | * locked by caller. If we are going to drop the final active reference, |
| 212 | * lock will be acquired prior to that. |
| 213 | */ |
| 214 | void deactivate_super(struct super_block *s) |
| 215 | { |
| 216 | if (!atomic_add_unless(&s->s_active, -1, 1)) { |
| 217 | down_write(&s->s_umount); |
| 218 | deactivate_locked_super(s); |
| 219 | } |
| 220 | } |
| 221 | |
| 222 | EXPORT_SYMBOL(deactivate_super); |
| 223 | |
| 224 | /** |
| 225 | * grab_super - acquire an active reference |
| 226 | * @s: reference we are trying to make active |
| 227 | * |
| 228 | * Tries to acquire an active reference. grab_super() is used when we |
| 229 | * had just found a superblock in super_blocks or fs_type->fs_supers |
| 230 | * and want to turn it into a full-blown active reference. grab_super() |
| 231 | * is called with sb_lock held and drops it. Returns 1 in case of |
| 232 | * success, 0 if we had failed (superblock contents was already dead or |
| 233 | * dying when grab_super() had been called). |
| 234 | */ |
| 235 | static int grab_super(struct super_block *s) __releases(sb_lock) |
| 236 | { |
| 237 | if (atomic_inc_not_zero(&s->s_active)) { |
| 238 | spin_unlock(&sb_lock); |
| 239 | down_write(&s->s_umount); |
| 240 | return 1; |
| 241 | } |
| 242 | /* it's going away */ |
| 243 | s->s_count++; |
| 244 | spin_unlock(&sb_lock); |
| 245 | /* wait for it to die */ |
| 246 | down_write(&s->s_umount); |
| 247 | up_write(&s->s_umount); |
| 248 | put_super(s); |
| 249 | return 0; |
| 250 | } |
| 251 | |
| 252 | /* |
| 253 | * Superblock locking. We really ought to get rid of these two. |
| 254 | */ |
| 255 | void lock_super(struct super_block * sb) |
| 256 | { |
| 257 | get_fs_excl(); |
| 258 | mutex_lock(&sb->s_lock); |
| 259 | } |
| 260 | |
| 261 | void unlock_super(struct super_block * sb) |
| 262 | { |
| 263 | put_fs_excl(); |
| 264 | mutex_unlock(&sb->s_lock); |
| 265 | } |
| 266 | |
| 267 | EXPORT_SYMBOL(lock_super); |
| 268 | EXPORT_SYMBOL(unlock_super); |
| 269 | |
| 270 | /** |
| 271 | * generic_shutdown_super - common helper for ->kill_sb() |
| 272 | * @sb: superblock to kill |
| 273 | * |
| 274 | * generic_shutdown_super() does all fs-independent work on superblock |
| 275 | * shutdown. Typical ->kill_sb() should pick all fs-specific objects |
| 276 | * that need destruction out of superblock, call generic_shutdown_super() |
| 277 | * and release aforementioned objects. Note: dentries and inodes _are_ |
| 278 | * taken care of and do not need specific handling. |
| 279 | * |
| 280 | * Upon calling this function, the filesystem may no longer alter or |
| 281 | * rearrange the set of dentries belonging to this super_block, nor may it |
| 282 | * change the attachments of dentries to inodes. |
| 283 | */ |
| 284 | void generic_shutdown_super(struct super_block *sb) |
| 285 | { |
| 286 | const struct super_operations *sop = sb->s_op; |
| 287 | |
| 288 | |
| 289 | if (sb->s_root) { |
| 290 | shrink_dcache_for_umount(sb); |
| 291 | sync_filesystem(sb); |
| 292 | get_fs_excl(); |
| 293 | sb->s_flags &= ~MS_ACTIVE; |
| 294 | |
| 295 | /* bad name - it should be evict_inodes() */ |
| 296 | invalidate_inodes(sb); |
| 297 | |
| 298 | if (sop->put_super) |
| 299 | sop->put_super(sb); |
| 300 | |
| 301 | /* Forget any remaining inodes */ |
| 302 | if (invalidate_inodes(sb)) { |
| 303 | printk("VFS: Busy inodes after unmount of %s. " |
| 304 | "Self-destruct in 5 seconds. Have a nice day...\n", |
| 305 | sb->s_id); |
| 306 | } |
| 307 | put_fs_excl(); |
| 308 | } |
| 309 | spin_lock(&sb_lock); |
| 310 | /* should be initialized for __put_super_and_need_restart() */ |
| 311 | list_del_init(&sb->s_list); |
| 312 | list_del(&sb->s_instances); |
| 313 | spin_unlock(&sb_lock); |
| 314 | up_write(&sb->s_umount); |
| 315 | } |
| 316 | |
| 317 | EXPORT_SYMBOL(generic_shutdown_super); |
| 318 | |
| 319 | /** |
| 320 | * sget - find or create a superblock |
| 321 | * @type: filesystem type superblock should belong to |
| 322 | * @test: comparison callback |
| 323 | * @set: setup callback |
| 324 | * @data: argument to each of them |
| 325 | */ |
| 326 | struct super_block *sget(struct file_system_type *type, |
| 327 | int (*test)(struct super_block *,void *), |
| 328 | int (*set)(struct super_block *,void *), |
| 329 | void *data) |
| 330 | { |
| 331 | struct super_block *s = NULL; |
| 332 | struct super_block *old; |
| 333 | int err; |
| 334 | |
| 335 | retry: |
| 336 | spin_lock(&sb_lock); |
| 337 | if (test) { |
| 338 | list_for_each_entry(old, &type->fs_supers, s_instances) { |
| 339 | if (!test(old, data)) |
| 340 | continue; |
| 341 | if (!grab_super(old)) |
| 342 | goto retry; |
| 343 | if (s) { |
| 344 | up_write(&s->s_umount); |
| 345 | destroy_super(s); |
| 346 | } |
| 347 | return old; |
| 348 | } |
| 349 | } |
| 350 | if (!s) { |
| 351 | spin_unlock(&sb_lock); |
| 352 | s = alloc_super(type); |
| 353 | if (!s) |
| 354 | return ERR_PTR(-ENOMEM); |
| 355 | goto retry; |
| 356 | } |
| 357 | |
| 358 | err = set(s, data); |
| 359 | if (err) { |
| 360 | spin_unlock(&sb_lock); |
| 361 | up_write(&s->s_umount); |
| 362 | destroy_super(s); |
| 363 | return ERR_PTR(err); |
| 364 | } |
| 365 | s->s_type = type; |
| 366 | strlcpy(s->s_id, type->name, sizeof(s->s_id)); |
| 367 | list_add_tail(&s->s_list, &super_blocks); |
| 368 | list_add(&s->s_instances, &type->fs_supers); |
| 369 | spin_unlock(&sb_lock); |
| 370 | get_filesystem(type); |
| 371 | return s; |
| 372 | } |
| 373 | |
| 374 | EXPORT_SYMBOL(sget); |
| 375 | |
| 376 | void drop_super(struct super_block *sb) |
| 377 | { |
| 378 | up_read(&sb->s_umount); |
| 379 | put_super(sb); |
| 380 | } |
| 381 | |
| 382 | EXPORT_SYMBOL(drop_super); |
| 383 | |
| 384 | /** |
| 385 | * sync_supers - helper for periodic superblock writeback |
| 386 | * |
| 387 | * Call the write_super method if present on all dirty superblocks in |
| 388 | * the system. This is for the periodic writeback used by most older |
| 389 | * filesystems. For data integrity superblock writeback use |
| 390 | * sync_filesystems() instead. |
| 391 | * |
| 392 | * Note: check the dirty flag before waiting, so we don't |
| 393 | * hold up the sync while mounting a device. (The newly |
| 394 | * mounted device won't need syncing.) |
| 395 | */ |
| 396 | void sync_supers(void) |
| 397 | { |
| 398 | struct super_block *sb; |
| 399 | |
| 400 | spin_lock(&sb_lock); |
| 401 | restart: |
| 402 | list_for_each_entry(sb, &super_blocks, s_list) { |
| 403 | if (sb->s_op->write_super && sb->s_dirt) { |
| 404 | sb->s_count++; |
| 405 | spin_unlock(&sb_lock); |
| 406 | |
| 407 | down_read(&sb->s_umount); |
| 408 | if (sb->s_root && sb->s_dirt) |
| 409 | sb->s_op->write_super(sb); |
| 410 | up_read(&sb->s_umount); |
| 411 | |
| 412 | spin_lock(&sb_lock); |
| 413 | if (__put_super_and_need_restart(sb)) |
| 414 | goto restart; |
| 415 | } |
| 416 | } |
| 417 | spin_unlock(&sb_lock); |
| 418 | } |
| 419 | |
| 420 | /** |
| 421 | * get_super - get the superblock of a device |
| 422 | * @bdev: device to get the superblock for |
| 423 | * |
| 424 | * Scans the superblock list and finds the superblock of the file system |
| 425 | * mounted on the device given. %NULL is returned if no match is found. |
| 426 | */ |
| 427 | |
| 428 | struct super_block * get_super(struct block_device *bdev) |
| 429 | { |
| 430 | struct super_block *sb; |
| 431 | |
| 432 | if (!bdev) |
| 433 | return NULL; |
| 434 | |
| 435 | spin_lock(&sb_lock); |
| 436 | rescan: |
| 437 | list_for_each_entry(sb, &super_blocks, s_list) { |
| 438 | if (sb->s_bdev == bdev) { |
| 439 | sb->s_count++; |
| 440 | spin_unlock(&sb_lock); |
| 441 | down_read(&sb->s_umount); |
| 442 | if (sb->s_root) |
| 443 | return sb; |
| 444 | up_read(&sb->s_umount); |
| 445 | /* restart only when sb is no longer on the list */ |
| 446 | spin_lock(&sb_lock); |
| 447 | if (__put_super_and_need_restart(sb)) |
| 448 | goto rescan; |
| 449 | } |
| 450 | } |
| 451 | spin_unlock(&sb_lock); |
| 452 | return NULL; |
| 453 | } |
| 454 | |
| 455 | EXPORT_SYMBOL(get_super); |
| 456 | |
| 457 | /** |
| 458 | * get_active_super - get an active reference to the superblock of a device |
| 459 | * @bdev: device to get the superblock for |
| 460 | * |
| 461 | * Scans the superblock list and finds the superblock of the file system |
| 462 | * mounted on the device given. Returns the superblock with an active |
| 463 | * reference and s_umount held exclusively or %NULL if none was found. |
| 464 | */ |
| 465 | struct super_block *get_active_super(struct block_device *bdev) |
| 466 | { |
| 467 | struct super_block *sb; |
| 468 | |
| 469 | if (!bdev) |
| 470 | return NULL; |
| 471 | |
| 472 | spin_lock(&sb_lock); |
| 473 | list_for_each_entry(sb, &super_blocks, s_list) { |
| 474 | if (sb->s_bdev != bdev) |
| 475 | continue; |
| 476 | |
| 477 | if (grab_super(sb)) /* drops sb_lock */ |
| 478 | return sb; |
| 479 | |
| 480 | spin_lock(&sb_lock); |
| 481 | } |
| 482 | spin_unlock(&sb_lock); |
| 483 | return NULL; |
| 484 | } |
| 485 | |
| 486 | struct super_block * user_get_super(dev_t dev) |
| 487 | { |
| 488 | struct super_block *sb; |
| 489 | |
| 490 | spin_lock(&sb_lock); |
| 491 | rescan: |
| 492 | list_for_each_entry(sb, &super_blocks, s_list) { |
| 493 | if (sb->s_dev == dev) { |
| 494 | sb->s_count++; |
| 495 | spin_unlock(&sb_lock); |
| 496 | down_read(&sb->s_umount); |
| 497 | if (sb->s_root) |
| 498 | return sb; |
| 499 | up_read(&sb->s_umount); |
| 500 | /* restart only when sb is no longer on the list */ |
| 501 | spin_lock(&sb_lock); |
| 502 | if (__put_super_and_need_restart(sb)) |
| 503 | goto rescan; |
| 504 | } |
| 505 | } |
| 506 | spin_unlock(&sb_lock); |
| 507 | return NULL; |
| 508 | } |
| 509 | |
| 510 | SYSCALL_DEFINE2(ustat, unsigned, dev, struct ustat __user *, ubuf) |
| 511 | { |
| 512 | struct super_block *s; |
| 513 | struct ustat tmp; |
| 514 | struct kstatfs sbuf; |
| 515 | int err = -EINVAL; |
| 516 | |
| 517 | s = user_get_super(new_decode_dev(dev)); |
| 518 | if (s == NULL) |
| 519 | goto out; |
| 520 | err = vfs_statfs(s->s_root, &sbuf); |
| 521 | drop_super(s); |
| 522 | if (err) |
| 523 | goto out; |
| 524 | |
| 525 | memset(&tmp,0,sizeof(struct ustat)); |
| 526 | tmp.f_tfree = sbuf.f_bfree; |
| 527 | tmp.f_tinode = sbuf.f_ffree; |
| 528 | |
| 529 | err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0; |
| 530 | out: |
| 531 | return err; |
| 532 | } |
| 533 | |
| 534 | /** |
| 535 | * do_remount_sb - asks filesystem to change mount options. |
| 536 | * @sb: superblock in question |
| 537 | * @flags: numeric part of options |
| 538 | * @data: the rest of options |
| 539 | * @force: whether or not to force the change |
| 540 | * |
| 541 | * Alters the mount options of a mounted file system. |
| 542 | */ |
| 543 | int do_remount_sb(struct super_block *sb, int flags, void *data, int force) |
| 544 | { |
| 545 | int retval; |
| 546 | int remount_rw, remount_ro; |
| 547 | |
| 548 | if (sb->s_frozen != SB_UNFROZEN) |
| 549 | return -EBUSY; |
| 550 | |
| 551 | #ifdef CONFIG_BLOCK |
| 552 | if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev)) |
| 553 | return -EACCES; |
| 554 | #endif |
| 555 | |
| 556 | if (flags & MS_RDONLY) |
| 557 | acct_auto_close(sb); |
| 558 | shrink_dcache_sb(sb); |
| 559 | sync_filesystem(sb); |
| 560 | |
| 561 | remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY); |
| 562 | remount_rw = !(flags & MS_RDONLY) && (sb->s_flags & MS_RDONLY); |
| 563 | |
| 564 | /* If we are remounting RDONLY and current sb is read/write, |
| 565 | make sure there are no rw files opened */ |
| 566 | if (remount_ro) { |
| 567 | if (force) |
| 568 | mark_files_ro(sb); |
| 569 | else if (!fs_may_remount_ro(sb)) |
| 570 | return -EBUSY; |
| 571 | retval = vfs_dq_off(sb, 1); |
| 572 | if (retval < 0 && retval != -ENOSYS) |
| 573 | return -EBUSY; |
| 574 | } |
| 575 | |
| 576 | if (sb->s_op->remount_fs) { |
| 577 | retval = sb->s_op->remount_fs(sb, &flags, data); |
| 578 | if (retval) |
| 579 | return retval; |
| 580 | } |
| 581 | sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK); |
| 582 | if (remount_rw) |
| 583 | vfs_dq_quota_on_remount(sb); |
| 584 | /* |
| 585 | * Some filesystems modify their metadata via some other path than the |
| 586 | * bdev buffer cache (eg. use a private mapping, or directories in |
| 587 | * pagecache, etc). Also file data modifications go via their own |
| 588 | * mappings. So If we try to mount readonly then copy the filesystem |
| 589 | * from bdev, we could get stale data, so invalidate it to give a best |
| 590 | * effort at coherency. |
| 591 | */ |
| 592 | if (remount_ro && sb->s_bdev) |
| 593 | invalidate_bdev(sb->s_bdev); |
| 594 | return 0; |
| 595 | } |
| 596 | |
| 597 | static void do_emergency_remount(struct work_struct *work) |
| 598 | { |
| 599 | struct super_block *sb; |
| 600 | |
| 601 | spin_lock(&sb_lock); |
| 602 | list_for_each_entry(sb, &super_blocks, s_list) { |
| 603 | sb->s_count++; |
| 604 | spin_unlock(&sb_lock); |
| 605 | down_write(&sb->s_umount); |
| 606 | if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) { |
| 607 | /* |
| 608 | * What lock protects sb->s_flags?? |
| 609 | */ |
| 610 | do_remount_sb(sb, MS_RDONLY, NULL, 1); |
| 611 | } |
| 612 | up_write(&sb->s_umount); |
| 613 | put_super(sb); |
| 614 | spin_lock(&sb_lock); |
| 615 | } |
| 616 | spin_unlock(&sb_lock); |
| 617 | kfree(work); |
| 618 | printk("Emergency Remount complete\n"); |
| 619 | } |
| 620 | |
| 621 | void emergency_remount(void) |
| 622 | { |
| 623 | struct work_struct *work; |
| 624 | |
| 625 | work = kmalloc(sizeof(*work), GFP_ATOMIC); |
| 626 | if (work) { |
| 627 | INIT_WORK(work, do_emergency_remount); |
| 628 | schedule_work(work); |
| 629 | } |
| 630 | } |
| 631 | |
| 632 | /* |
| 633 | * Unnamed block devices are dummy devices used by virtual |
| 634 | * filesystems which don't use real block-devices. -- jrs |
| 635 | */ |
| 636 | |
| 637 | static DEFINE_IDA(unnamed_dev_ida); |
| 638 | static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */ |
| 639 | static int unnamed_dev_start = 0; /* don't bother trying below it */ |
| 640 | |
| 641 | int set_anon_super(struct super_block *s, void *data) |
| 642 | { |
| 643 | int dev; |
| 644 | int error; |
| 645 | |
| 646 | retry: |
| 647 | if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0) |
| 648 | return -ENOMEM; |
| 649 | spin_lock(&unnamed_dev_lock); |
| 650 | error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev); |
| 651 | if (!error) |
| 652 | unnamed_dev_start = dev + 1; |
| 653 | spin_unlock(&unnamed_dev_lock); |
| 654 | if (error == -EAGAIN) |
| 655 | /* We raced and lost with another CPU. */ |
| 656 | goto retry; |
| 657 | else if (error) |
| 658 | return -EAGAIN; |
| 659 | |
| 660 | if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) { |
| 661 | spin_lock(&unnamed_dev_lock); |
| 662 | ida_remove(&unnamed_dev_ida, dev); |
| 663 | if (unnamed_dev_start > dev) |
| 664 | unnamed_dev_start = dev; |
| 665 | spin_unlock(&unnamed_dev_lock); |
| 666 | return -EMFILE; |
| 667 | } |
| 668 | s->s_dev = MKDEV(0, dev & MINORMASK); |
| 669 | s->s_bdi = &noop_backing_dev_info; |
| 670 | return 0; |
| 671 | } |
| 672 | |
| 673 | EXPORT_SYMBOL(set_anon_super); |
| 674 | |
| 675 | void kill_anon_super(struct super_block *sb) |
| 676 | { |
| 677 | int slot = MINOR(sb->s_dev); |
| 678 | |
| 679 | generic_shutdown_super(sb); |
| 680 | spin_lock(&unnamed_dev_lock); |
| 681 | ida_remove(&unnamed_dev_ida, slot); |
| 682 | if (slot < unnamed_dev_start) |
| 683 | unnamed_dev_start = slot; |
| 684 | spin_unlock(&unnamed_dev_lock); |
| 685 | } |
| 686 | |
| 687 | EXPORT_SYMBOL(kill_anon_super); |
| 688 | |
| 689 | void kill_litter_super(struct super_block *sb) |
| 690 | { |
| 691 | if (sb->s_root) |
| 692 | d_genocide(sb->s_root); |
| 693 | kill_anon_super(sb); |
| 694 | } |
| 695 | |
| 696 | EXPORT_SYMBOL(kill_litter_super); |
| 697 | |
| 698 | static int ns_test_super(struct super_block *sb, void *data) |
| 699 | { |
| 700 | return sb->s_fs_info == data; |
| 701 | } |
| 702 | |
| 703 | static int ns_set_super(struct super_block *sb, void *data) |
| 704 | { |
| 705 | sb->s_fs_info = data; |
| 706 | return set_anon_super(sb, NULL); |
| 707 | } |
| 708 | |
| 709 | int get_sb_ns(struct file_system_type *fs_type, int flags, void *data, |
| 710 | int (*fill_super)(struct super_block *, void *, int), |
| 711 | struct vfsmount *mnt) |
| 712 | { |
| 713 | struct super_block *sb; |
| 714 | |
| 715 | sb = sget(fs_type, ns_test_super, ns_set_super, data); |
| 716 | if (IS_ERR(sb)) |
| 717 | return PTR_ERR(sb); |
| 718 | |
| 719 | if (!sb->s_root) { |
| 720 | int err; |
| 721 | sb->s_flags = flags; |
| 722 | err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0); |
| 723 | if (err) { |
| 724 | deactivate_locked_super(sb); |
| 725 | return err; |
| 726 | } |
| 727 | |
| 728 | sb->s_flags |= MS_ACTIVE; |
| 729 | } |
| 730 | |
| 731 | simple_set_mnt(mnt, sb); |
| 732 | return 0; |
| 733 | } |
| 734 | |
| 735 | EXPORT_SYMBOL(get_sb_ns); |
| 736 | |
| 737 | #ifdef CONFIG_BLOCK |
| 738 | static int set_bdev_super(struct super_block *s, void *data) |
| 739 | { |
| 740 | s->s_bdev = data; |
| 741 | s->s_dev = s->s_bdev->bd_dev; |
| 742 | |
| 743 | /* |
| 744 | * We set the bdi here to the queue backing, file systems can |
| 745 | * overwrite this in ->fill_super() |
| 746 | */ |
| 747 | s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info; |
| 748 | return 0; |
| 749 | } |
| 750 | |
| 751 | static int test_bdev_super(struct super_block *s, void *data) |
| 752 | { |
| 753 | return (void *)s->s_bdev == data; |
| 754 | } |
| 755 | |
| 756 | int get_sb_bdev(struct file_system_type *fs_type, |
| 757 | int flags, const char *dev_name, void *data, |
| 758 | int (*fill_super)(struct super_block *, void *, int), |
| 759 | struct vfsmount *mnt) |
| 760 | { |
| 761 | struct block_device *bdev; |
| 762 | struct super_block *s; |
| 763 | fmode_t mode = FMODE_READ; |
| 764 | int error = 0; |
| 765 | |
| 766 | if (!(flags & MS_RDONLY)) |
| 767 | mode |= FMODE_WRITE; |
| 768 | |
| 769 | bdev = open_bdev_exclusive(dev_name, mode, fs_type); |
| 770 | if (IS_ERR(bdev)) |
| 771 | return PTR_ERR(bdev); |
| 772 | |
| 773 | /* |
| 774 | * once the super is inserted into the list by sget, s_umount |
| 775 | * will protect the lockfs code from trying to start a snapshot |
| 776 | * while we are mounting |
| 777 | */ |
| 778 | mutex_lock(&bdev->bd_fsfreeze_mutex); |
| 779 | if (bdev->bd_fsfreeze_count > 0) { |
| 780 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 781 | error = -EBUSY; |
| 782 | goto error_bdev; |
| 783 | } |
| 784 | s = sget(fs_type, test_bdev_super, set_bdev_super, bdev); |
| 785 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
| 786 | if (IS_ERR(s)) |
| 787 | goto error_s; |
| 788 | |
| 789 | if (s->s_root) { |
| 790 | if ((flags ^ s->s_flags) & MS_RDONLY) { |
| 791 | deactivate_locked_super(s); |
| 792 | error = -EBUSY; |
| 793 | goto error_bdev; |
| 794 | } |
| 795 | |
| 796 | close_bdev_exclusive(bdev, mode); |
| 797 | } else { |
| 798 | char b[BDEVNAME_SIZE]; |
| 799 | |
| 800 | s->s_flags = flags; |
| 801 | s->s_mode = mode; |
| 802 | strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id)); |
| 803 | sb_set_blocksize(s, block_size(bdev)); |
| 804 | error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); |
| 805 | if (error) { |
| 806 | deactivate_locked_super(s); |
| 807 | goto error; |
| 808 | } |
| 809 | |
| 810 | s->s_flags |= MS_ACTIVE; |
| 811 | bdev->bd_super = s; |
| 812 | } |
| 813 | |
| 814 | simple_set_mnt(mnt, s); |
| 815 | return 0; |
| 816 | |
| 817 | error_s: |
| 818 | error = PTR_ERR(s); |
| 819 | error_bdev: |
| 820 | close_bdev_exclusive(bdev, mode); |
| 821 | error: |
| 822 | return error; |
| 823 | } |
| 824 | |
| 825 | EXPORT_SYMBOL(get_sb_bdev); |
| 826 | |
| 827 | void kill_block_super(struct super_block *sb) |
| 828 | { |
| 829 | struct block_device *bdev = sb->s_bdev; |
| 830 | fmode_t mode = sb->s_mode; |
| 831 | |
| 832 | bdev->bd_super = NULL; |
| 833 | generic_shutdown_super(sb); |
| 834 | sync_blockdev(bdev); |
| 835 | close_bdev_exclusive(bdev, mode); |
| 836 | } |
| 837 | |
| 838 | EXPORT_SYMBOL(kill_block_super); |
| 839 | #endif |
| 840 | |
| 841 | int get_sb_nodev(struct file_system_type *fs_type, |
| 842 | int flags, void *data, |
| 843 | int (*fill_super)(struct super_block *, void *, int), |
| 844 | struct vfsmount *mnt) |
| 845 | { |
| 846 | int error; |
| 847 | struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL); |
| 848 | |
| 849 | if (IS_ERR(s)) |
| 850 | return PTR_ERR(s); |
| 851 | |
| 852 | s->s_flags = flags; |
| 853 | |
| 854 | error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); |
| 855 | if (error) { |
| 856 | deactivate_locked_super(s); |
| 857 | return error; |
| 858 | } |
| 859 | s->s_flags |= MS_ACTIVE; |
| 860 | simple_set_mnt(mnt, s); |
| 861 | return 0; |
| 862 | } |
| 863 | |
| 864 | EXPORT_SYMBOL(get_sb_nodev); |
| 865 | |
| 866 | static int compare_single(struct super_block *s, void *p) |
| 867 | { |
| 868 | return 1; |
| 869 | } |
| 870 | |
| 871 | int get_sb_single(struct file_system_type *fs_type, |
| 872 | int flags, void *data, |
| 873 | int (*fill_super)(struct super_block *, void *, int), |
| 874 | struct vfsmount *mnt) |
| 875 | { |
| 876 | struct super_block *s; |
| 877 | int error; |
| 878 | |
| 879 | s = sget(fs_type, compare_single, set_anon_super, NULL); |
| 880 | if (IS_ERR(s)) |
| 881 | return PTR_ERR(s); |
| 882 | if (!s->s_root) { |
| 883 | s->s_flags = flags; |
| 884 | error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); |
| 885 | if (error) { |
| 886 | deactivate_locked_super(s); |
| 887 | return error; |
| 888 | } |
| 889 | s->s_flags |= MS_ACTIVE; |
| 890 | } else { |
| 891 | do_remount_sb(s, flags, data, 0); |
| 892 | } |
| 893 | simple_set_mnt(mnt, s); |
| 894 | return 0; |
| 895 | } |
| 896 | |
| 897 | EXPORT_SYMBOL(get_sb_single); |
| 898 | |
| 899 | struct vfsmount * |
| 900 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) |
| 901 | { |
| 902 | struct vfsmount *mnt; |
| 903 | char *secdata = NULL; |
| 904 | int error; |
| 905 | |
| 906 | if (!type) |
| 907 | return ERR_PTR(-ENODEV); |
| 908 | |
| 909 | error = -ENOMEM; |
| 910 | mnt = alloc_vfsmnt(name); |
| 911 | if (!mnt) |
| 912 | goto out; |
| 913 | |
| 914 | if (flags & MS_KERNMOUNT) |
| 915 | mnt->mnt_flags = MNT_INTERNAL; |
| 916 | |
| 917 | if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) { |
| 918 | secdata = alloc_secdata(); |
| 919 | if (!secdata) |
| 920 | goto out_mnt; |
| 921 | |
| 922 | error = security_sb_copy_data(data, secdata); |
| 923 | if (error) |
| 924 | goto out_free_secdata; |
| 925 | } |
| 926 | |
| 927 | error = type->get_sb(type, flags, name, data, mnt); |
| 928 | if (error < 0) |
| 929 | goto out_free_secdata; |
| 930 | BUG_ON(!mnt->mnt_sb); |
| 931 | WARN_ON(!mnt->mnt_sb->s_bdi); |
| 932 | |
| 933 | error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata); |
| 934 | if (error) |
| 935 | goto out_sb; |
| 936 | |
| 937 | /* |
| 938 | * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE |
| 939 | * but s_maxbytes was an unsigned long long for many releases. Throw |
| 940 | * this warning for a little while to try and catch filesystems that |
| 941 | * violate this rule. This warning should be either removed or |
| 942 | * converted to a BUG() in 2.6.34. |
| 943 | */ |
| 944 | WARN((mnt->mnt_sb->s_maxbytes < 0), "%s set sb->s_maxbytes to " |
| 945 | "negative value (%lld)\n", type->name, mnt->mnt_sb->s_maxbytes); |
| 946 | |
| 947 | mnt->mnt_mountpoint = mnt->mnt_root; |
| 948 | mnt->mnt_parent = mnt; |
| 949 | up_write(&mnt->mnt_sb->s_umount); |
| 950 | free_secdata(secdata); |
| 951 | return mnt; |
| 952 | out_sb: |
| 953 | dput(mnt->mnt_root); |
| 954 | deactivate_locked_super(mnt->mnt_sb); |
| 955 | out_free_secdata: |
| 956 | free_secdata(secdata); |
| 957 | out_mnt: |
| 958 | free_vfsmnt(mnt); |
| 959 | out: |
| 960 | return ERR_PTR(error); |
| 961 | } |
| 962 | |
| 963 | EXPORT_SYMBOL_GPL(vfs_kern_mount); |
| 964 | |
| 965 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
| 966 | { |
| 967 | int err; |
| 968 | const char *subtype = strchr(fstype, '.'); |
| 969 | if (subtype) { |
| 970 | subtype++; |
| 971 | err = -EINVAL; |
| 972 | if (!subtype[0]) |
| 973 | goto err; |
| 974 | } else |
| 975 | subtype = ""; |
| 976 | |
| 977 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); |
| 978 | err = -ENOMEM; |
| 979 | if (!mnt->mnt_sb->s_subtype) |
| 980 | goto err; |
| 981 | return mnt; |
| 982 | |
| 983 | err: |
| 984 | mntput(mnt); |
| 985 | return ERR_PTR(err); |
| 986 | } |
| 987 | |
| 988 | struct vfsmount * |
| 989 | do_kern_mount(const char *fstype, int flags, const char *name, void *data) |
| 990 | { |
| 991 | struct file_system_type *type = get_fs_type(fstype); |
| 992 | struct vfsmount *mnt; |
| 993 | if (!type) |
| 994 | return ERR_PTR(-ENODEV); |
| 995 | mnt = vfs_kern_mount(type, flags, name, data); |
| 996 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && |
| 997 | !mnt->mnt_sb->s_subtype) |
| 998 | mnt = fs_set_subtype(mnt, fstype); |
| 999 | put_filesystem(type); |
| 1000 | return mnt; |
| 1001 | } |
| 1002 | EXPORT_SYMBOL_GPL(do_kern_mount); |
| 1003 | |
| 1004 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) |
| 1005 | { |
| 1006 | return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); |
| 1007 | } |
| 1008 | |
| 1009 | EXPORT_SYMBOL_GPL(kern_mount_data); |